Discussion Radeon 6500XT and 6400

Page 3 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

Insert_Nickname

Diamond Member
May 6, 2012
4,582
1,164
136
16CU N24 is supposedly ~141 mm^2 built using 6nm process, Videocardz.

If we compare It to 24CU N14 with a die size of 158 mm^2, then It doesn't look so good, 1/3 less CU, 1/2 memory controller, +16MB IC and you save only 10% by using a better process.
To have the same TFLOPs as a fully unlocked N14, you would need to have 50% higher clocks, in other words ~2.7-2.8GHz. Against RX 5500XT you only need ~38% higher clocks, in other words ~2.45-2.55GHz.
I don't think 2.6'ish GHz is out of the question. Perhaps even 2.7-2.8. My 6600XT routinely runs at 2.6+ GHz, so it isn't unfeasible.

Another problem is only 4GB Vram and at 1440p only 64bit GDDR6 will also be a bottleneck, we can also see this with RX 6600XT at 4K.
The only real positive about N24 should be the power consumption, hopefully It will be <=75W.
These cards aren't targeted at 1440p. The 6600XT would be my minimum for that. In fact it is my minimum. 4K? 6800 as a minimum. But I'm not taking out a mortgage for one of those.

With regards to bandwidth, don't underestimate that Infinity Cache, and assuming 16GHz GDDR6 @ 64bit is still 128GB/s of bandwidth. Upping to the newer 20GHz would be 160GB/s. Samsung actually makes 24GHz capable GDDR6, so 192GB/s is theoretically possible. Should be plenty for 1080p. Even if we're talking GTX1060 class performance, that's still a competent 1080p card. Of course, you have to maybe turn down a setting or two, and forget about ray tracing, but should still be good.
 
  • Like
Reactions: dvsv

TESKATLIPOKA

Senior member
May 1, 2020
568
603
106
I don't think 2.6'ish GHz is out of the question. Perhaps even 2.7-2.8. My 6600XT routinely runs at 2.6+ GHz, so it isn't unfeasible.
I think It can achieve such clockspeed, but in my opinion the problem is that you need such a high clockspeed just to be on par with the previous generation.
In my opinion they should have added +4CU for 20CU in total. Yes, die size would increase by ~8-10mm^2, but performance would also increase by ~20%.

These cards aren't targeted at 1440p. The 6600XT would be my minimum for that. In fact it is my minimum. 4K? 6800 as a minimum. But I'm not taking out a mortgage for one of those.

With regards to bandwidth, don't underestimate that Infinity Cache, and assuming 16GHz GDDR6 @ 64bit is still 128GB/s of bandwidth. Upping to the newer 20GHz would be 160GB/s. Samsung actually makes 24GHz capable GDDR6, so 192GB/s is theoretically possible. Should be plenty for 1080p. Even if we're talking GTX1060 class performance, that's still a competent 1080p card. Of course, you have to maybe turn down a setting or two, and forget about ray tracing, but should still be good.
I know N24 is aimed for Full HD.
What I wanted to say was that If It performed the same as RX 5500XT at Full HD, then at higher resolutions It would start to lose against It. The same thing you can see with RX 6600XT vs RX 5700XT. Infinity Cache is an interesting idea, but It has Its advantages and disadvantages.
 
Last edited:
  • Like
Reactions: Insert_Nickname

Glo.

Diamond Member
Apr 25, 2015
4,973
3,588
136
The only real positive about N24 should be the power consumption, hopefully It will be <=75W.
Its 107W TGP for 6500 XT and 53W's for 6400.

The GPU die is cut by 50% compared to 6600 XT but TGP is only cut by around 30-35%.
 

TESKATLIPOKA

Senior member
May 1, 2020
568
603
106
Its 107W TGP for 6500 XT and 53W's for 6400.

The GPU die is cut by 50% compared to 6600 XT but TGP is only cut by around 30-35%.
I saw that on the web, but honestly I don't believe It. The difference as far as I know is 12CU vs 16CU(+33%) and higher clocks, yet It has 2x higher TGP?
 

Glo.

Diamond Member
Apr 25, 2015
4,973
3,588
136
I saw that on the web, but honestly I don't believe It. The difference as far as I know is 12CU vs 16CU(+33%) and higher clocks, yet It has 2x higher TGP?
Its clocked to hell. Higher than 6600 XT. Expect 2.6 GHz game clock, at least, and at least GTX 1660 non-super perf.
 

Glo.

Diamond Member
Apr 25, 2015
4,973
3,588
136
I think It can achieve such clockspeed, but in my opinion the problem is that you need such a high clockspeed just to be on par with the previous generation.
In my opinion they should have added +4CU for 20CU in total. Yes, die size would increase by ~8-10mm^2, but performance would also increase by ~20%.
Memory bandwidth would be required to be higher, and it would require to redesign the layout. It appears that N24 is single Shader Engine and Shader Array. 1280 ALUs would require at least 96 bit bus, and dual shader engine. It would increase the die size by much more than just 8-10 mm2.
 

DeathReborn

Platinum Member
Oct 11, 2005
2,458
375
126
Its clocked to hell. Higher than 6600 XT. Expect 2.6 GHz game clock, at least, and at least GTX 1660 non-super perf.
I am expecting maybe 10-15% faster than the 5500XT, but it will struggle at 1080p in some titles when you crank up the settings to High/ultra due to the memory.
 
  • Like
Reactions: Glo.

TESKATLIPOKA

Senior member
May 1, 2020
568
603
106
Its clocked to hell. Higher than 6600 XT. Expect 2.6 GHz game clock, at least, and at least GTX 1660 non-super perf.
It is still very high, If you consider a 28(32)CU RX 6600(XT) 8GB has 132(160)W TBP and is built using an older node.

Memory bandwidth would be required to be higher, and it would require to redesign the layout. It appears that N24 is single Shader Engine and Shader Array. 1280 ALUs would require at least 96 bit bus, and dual shader engine. It would increase the die size by much more than just 8-10 mm2.
20CU N24 with a single Shader engine would be exactly 1/4 of N21 and that one has 4 shader Engines and 80CU, so that shouldn't be a problem.
Bandwidth could be an issue, true.
 
Last edited:

jpiniero

Lifer
Oct 1, 2010
11,320
3,052
136
Only thing I can think of is the patent costs were a big enough deal when Navi 24 was assumed to be 100, 150 bucks if released that they decided to not include the block.
 

eek2121

Golden Member
Aug 2, 2005
1,800
1,982
136
16CU N24 is supposedly ~141 mm^2 built using 6nm process, Videocardz.

If we compare It to 24CU N14 with a die size of 158 mm^2, then It doesn't look very good -> 1/3 less CU, 1/2 memory controller, +16MB IC and you save only 10% by using a better process.
To have the same TFLOPs as a fully unlocked N14, you would need to have 50% higher clocks, in other words ~2.7-2.8GHz. Against RX 5500XT you only need ~38% higher clocks, in other words ~2.45-2.55GHz.
Another problem is only 4GB Vram and at 1440p only 64bit GDDR6 will also be a bottleneck, we can also see this with RX 6600XT at 4K.
The only real positive about N24 should be the power consumption, hopefully It will be <=75W.
N6 wasn’t providing much of a shrink (theoretical max 18%) to begin with. Also these cards are optimized for high clocks, which means AMD is sacrificing density.
 

Mopetar

Diamond Member
Jan 31, 2011
6,677
3,720
136
N6 wasn’t providing much of a shrink (theoretical max 18%) to begin with. Also these cards are optimized for high clocks, which means AMD is sacrificing density.
Don't you always get the shrink (at least to the extent possible, which usually refers to logic) and the trade-off is either lower operating power at the same frequency or higher frequency at the same operating power?

I know that you can make some trade-offs such as keeping some dark silicon around hot spots to help dissipate the heat, which is trading density for performance, but I have no idea to what extend AMD would be making those kinds of decisions on an early product for a node. It seems (to me at least) like N24 is a bit of a pipe cleaner product to learn those kinds of lessons for something made later on that same node.
 

Glo.

Diamond Member
Apr 25, 2015
4,973
3,588
136
IMO, in games that do not exceed 4 GB VRAM you will get GTX 1660 Ti performance levels.
 

GodisanAtheist

Diamond Member
Nov 16, 2006
3,932
2,474
136
RX 6500XT:
boost clock: 2815 MHz
game clock: 2610 MHz
Videocardz

Memory used is 18GHz according to greymon55.

P.S. He also mentioned, that RX 6400 is supposedly available only for OEMs.
-Those are some beefy clocks for both the core and the RAM for a low end card. I fully expect vendor models to run the gamut on memory though, suspect we'll see plenty of crap tier ram on these cards also without any distinction on the box.

Hopefully AMD lets vendors get away with that kind of thing only if it drops the XT moniker.
 

VirtualLarry

No Lifer
Aug 25, 2001
53,267
7,701
126
P.S. He also mentioned, that RX 6400 is supposedly available only for OEMs.
If the RX 6400 is the only one of the bunch to not require PCI-E power, then I can see why OEMs would want an allocation of them, but by the same token, they could make the GTX 1050 ti 4GB cards finally obsolete.
 

TESKATLIPOKA

Senior member
May 1, 2020
568
603
106
A leak about Radeon 6000 mobile GPUs.
Videocardz



For some reason this slide and the next one shows 128bit GDDR6 memory controller for N24 based RX 6300M and RX 6500M.
Most likely Typo, or there should be no reason for 18GHz memory.

RX 6300M has only 8MB of IC and only 1512MHz? If this low clockspeed is true, then even R7 6800U could end up faster depending on TDP. It has 12CU and 2200Mhz max boost and TDP 15-28W.

BTW here we can see how inefficient RX 6500XT really is with 2610Mhz gaming clockspeed. That's only 19% higher than RX 6500M with 2191Mhz, but It needs >2x more TGP.
 
Last edited:

ASK THE COMMUNITY